If an AI system can’t link your name, it can’t find you. And if it can’t find you, it will act like you don’t exist.
That is why real companies show up as “unknown” in LLM answers. Not because they are small. Not because the model is “bad.” But because the company’s name is scattered across the internet as messy variants that never get merged into one clean, canonical (accurate source of truth) entity.
This is the executive playbook for fixing that.
The quiet way brands disappear
Picture a buyer asking an LLM:
- “Who are the top vendors for X?”
- “Is Product Z legit?”
- “Compare A vs B.”
Now picture your company is mentioned across the web as:
- “Acme AI”
- “Acme-AI”
- “ACME.ai”
- “AcmeAI Inc.”
To a human, that’s obviously one company.
To many AI pipelines, it’s not.
They don’t “figure it out” like a person. They match strings. They link evidence. They cluster mentions. When the strings don’t match, the evidence doesn’t stack. And when the evidence doesn’t stack, you don’t rank.
So you get left out. Or you get blended into a competitor. Or the model shrugs and says it can’t find enough information.
That’s not a marketing problem. That’s an identity problem.
What “Name Hygiene” means (in plain terms)
Name hygiene is keeping one canonical form of your name and your key identifiers consistent everywhere AI systems read from.
It includes:
- Spelling: one spelling, always.
- Punctuation: pick “Acme AI” or “Acme.AI” or “Acme-AI.” Don’t rotate.
- Capitalization: “FlowKit” and “flowkit” are often treated as different strings.
- Suffixes: “Inc,” “LLC,” “Ltd” should be used on purpose, not randomly.
- Domains and handles: your domain and social handles should reinforce the same identity.
- Legacy names: rebrands create “ghost names” that keep splitting your presence for years.
what the world says → what you want AI to recognize.
Why a C-level leader should care
AI answers are becoming a distribution channel. Sometimes they’re the first channel.
If your name is hard for AI to resolve, you can lose in three ways:
- You don’t show up.
- You show up inconsistently (only for some spellings, some prompts, some users).
- You show up wrong (confused with someone else, wrong domain, wrong product, wrong facts).
Name hygiene is not “brand polish.” It’s infrastructure like DNS, identity, and billing. If it’s messy, everything above it becomes unreliable.
How AI systems “see” names (what actually breaks)
Most AI visibility stacks LLMs, search layers feeding LLMs, “AI overview” features follow a pattern like this:
- Extract mentions
Systems scan text and pull out “things that look like entities”: companies, products, people. - Link mentions to an entity record
They try to decide: “This string is that company.” - Aggregate evidence
They pool signals from many sources: citations, descriptions, web pages, profiles, directories. - Retrieve for a question
When someone asks a question, they pull the most relevant entities and evidence. - Answer + attribute
The model generates the answer and often names sources.
Name variants break Step 2. And Step 2 is the hinge. If linking breaks, your evidence never piles up into one strong, confident entity.
That’s how a real company becomes “unknown.”
The core failure: evidence fragmentation
Think of your credibility in an AI system as a pile of evidence:
- consistent mentions
- consistent domain links
- consistent descriptions
- consistent leadership/product associations
- strong citations from trusted sources
When your name is split, that evidence is split.
Instead of one strong pile, you get several weak piles. Weak piles don’t win retrieval. Weak piles don’t get cited. Weak piles get ignored.
This is also why “we got mentioned on a big site” sometimes doesn’t help: if the big site uses a different name variant, it may strengthen the wrong pile.
How much does being inconsistent costs you
One company. Three spellings. Three different outcomes.
Example:
| Name in the wild | Canonical name | Entity Score (wild) | Entity Score (canonical) | Question Score (wild) | Question Score (canonical) |
|---|---|---|---|---|---|
| “Acme-AI” | “Acme AI” | 41 | 72 | 28 | 66 |
| “ACME.ai” | “Acme AI” | 55 | 72 | 44 | 66 |
| “AcmeAI Inc.” | “Acme AI” | 36 | 72 | 19 | 66 |
Same business. Same reality. Different string. Different AI behavior.
That’s the story you’re telling: naming alone can create the appearance of being “unknown.
The only math most executives need (and why it matters)
You don’t need pages of formulas. You need one idea:
Your “real” score is the score you get where people actually mention you.
If most mentions are messy variants, your effective visibility behaves like the variants not like the canonical name you prefer.
- If 60% of your mentions use the “wrong” string, then 60% of the time the AI system starts from the wrong pile of evidence.
A simple back-of-the-napkin example:
- Canonical Q Score: 66
- Two common variants score lower: 44 and 28
- Mentions are split: 40% canonical, 35% variant A, 25% variant B
Your effective Q Score is roughly:
- (0.40 × 66) + (0.35 × 44) + (0.25 × 28) = 26.4 + 15.4 + 7.0 = 48.8
So even though your canonical form looks strong, the world experiences you closer to 49.
That gap is called name hygiene debt.
The four naming problems that cause most AI invisibility
1) Split clusters (the silent killer)
Your company becomes multiple “almost you” entities.
What it looks like
- The canonical name scores fine.
- Variants score weak.
- AI answers change based on spelling.
What to do
- Map variants → canonical (NormalizationPairs).
- Fix your owned surfaces so future mentions converge.
2) Collisions (you get merged into someone else)
Short or common names collide with other entities.
What it looks like
- You show up, but with the wrong facts.
- Wrong domain, wrong founder, wrong product.
What to do
- Add disambiguation signals everywhere: domain, category phrase, HQ, founder name, product descriptor.
- Make sure those signals are consistent on high-trust sources.
3) Rebrands (your past keeps splitting your present)
Old names live forever in PDFs, partner blogs, speaker bios, and directories.
What it looks like
- Two “official” names exist online.
- AI answers mix old and new.
- Scores wobble over time.
What to do
- redirects,
- updated boilerplates,
- partner correction outreach,
- a temporary “formerly known as” line (used consistently).
4) Domain confusion (AI can’t tell what your real “home” is)
You’re cited through resellers, old domains, or many different link targets.
What it looks like
- Mentions don’t point back to one primary domain.
- AI answers hedge: “might be,” “not sure,” or pick the wrong site.
What to do
- One primary domain.
- Redirect the rest.
- Make the primary domain the default link in press kits and partner pages.
The Name Hygiene Playbook (what to do, in order)
This is the part your team can execute.
Step 1: Choose one canonical identity (and write it down)
Decide, exactly:
- Canonical public name (spelling + caps + punctuation)
- Primary domain
- Product naming rules (especially “AI” features, model names, agents)
- What to do with legal suffixes (often: keep legal name for contracts, not public mentions)
This becomes policy. Not preference.
Step 2: Build the “alias firewall” (your NormalizationPairs sheet)
Your sheet should track, at minimum:
- Variant string
- Canonical string
- Where it appears (source)
- How often it appears (or a rough tier: high/med/low)
- Severity (does it cause splits or collisions?)
- Entity Score and Question Score for the variant vs canonical (use LLMtel.com)
Goal: not perfection. Goal: coverage of the variants that actually move your scores.
Step 3: Fix owned surfaces first (high leverage, low cost)
Start where you control the copy:
- Home page title + H1
- About page + boilerplate paragraph
- Press kit (this is a major identity source)
- Leadership bios (consistent company name + primary domain)
- Organization schema / structured metadata (if you publish it)
- Footer/copyright (don’t accidentally introduce a second “official” string)
- Downloadable PDFs (they get indexed, shared, and scraped for years)
Rule: every owned surface should repeat the canonical string and primary domain.
Step 4: Fix semi-owned surfaces (high trust, editable)
These often carry more authority than your blog:
- LinkedIn company page
- App store listings (if relevant)
- GitHub org name + description
- YouTube channel
- Major directories and review sites where you have access
If these disagree with your site, many systems trust them more than you.
Step 5: Fix earned surfaces (the web still talks about you)
You don’t need to fix the whole internet. You fix the parts that matter most:
- Top partner pages that rank
- Conference speaker pages
- Podcast show notes
- Old press release syndications that still show up in search
Pick the top sources that create the most damaging variants then work down the list.
How to prioritize, without turning this into a forever project
Use a simple rule:
- Fix the name variants that (1) show up often and (2) score much lower than your canonical name.
In practice, your team can rank variants with two columns:
- “How common is it?”
- “How big is the score gap?”
The top-right corner (common + big gap) is your first week of work.
What to measure (a clean executive dashboard)
For your company name, top product names, and key exec names, track:
- Canonical Entity Score and Q Score
- Top variants and their scores
- Share of mentions that are non-canonical (even rough tiers help)
- The “effective” score trend (are we converging toward canonical over time?)
- Collision checks (are we being confused with anyone else?)
his turns name hygiene from a brand debate into an operating metric.
Governance: how to keep it clean after you fix it
Most teams clean this once, then drift back into chaos through normal growth: new partners, new decks, new listings, new product pages, a new PR agency.
Three controls prevent drift:
1) A one-page Name Hygiene Spec
Include:
- Canonical company name (exact)
- Canonical product names (exact)
- Primary domain
- Allowed aliases (short list)
- Forbidden variants (long list)
- Standard boilerplate paragraph (copy/paste)
2) Put name hygiene into publishing and partnerships
Add a check to:
- press releases
- guest posts
- conference bios
- partner listings
- analyst briefings
- sales decks
3) Treat rebrands as migrations, every time
Rebrands without redirects and partner updates create permanent splits.
What “good” looks like
When name hygiene is working:
- Most mentions use the canonical string.
- Variants still exist, but they resolve back to the canonical entity.
- Your domain is consistently treated as the “home.”
- Your leadership and products stay attached to the right entity.
- LLM answers become boring in the best way: consistent, correct, repeatable.
Closing: naming is a prerequisite, not a tactic
If your name is not consistent, AI systems cannot stack evidence. If evidence can’t stack, you won’t rank. If you don’t rank, you don’t exist in the answers.
The good news: this is fixable, and your own data already shows the lift. Your NormalizationPairs sheet and the Entity Score / Q Score deltas are the story.
Name hygiene is how you turn “unknown” into “obvious.”